Agentic AI in the enterprise: Why leaders are rethinking trust and execution

Agentic AI in the enterprise 

Enterprise leaders are rethinking trust, data strategy, and organisational readiness for intelligent agents.

Even as recently as 2025 executive conversations about AI often carried the atmosphere of a gold rush. Leaders spoke in sweeping terms about transformation, productivity gains, and competitive reinvention. The dominant concern, as HotTopics heard across its C-suite debates, was speed; how quickly could organisations adopt these technologies before rivals pulled ahead?

In 2026 the mood has shifted perceptibly—and with it, executive priorities. This article reports on these expectations as they were put forward and debated during a recent Food for Thought roundtable at HotTopics’ The Studio, in partnership with Tredence, and Data Bricks.

Why agentic AI is different from earlier enterprise AI

Agentic AI are systems capable of generating outputs, certainly, but they are noted for initiating and coordinating actions, too. They dominate boardroom discussions as CEOs rapidly pivot to AI-first structures and cultures, often demanding their C-suites keep up with their pace and vision. Those executives, such as the CIO, CTO, Chief Data Officer, and CISO, have been some of the first to test, experiment, and lead with these new systems. But after several years of blockbuster investments, capital expenditure, and promises, the debate around AI has matured.

Indeed, around the table, technology executives have grown circumspect, careful even, but rigorous with this still-nascent innovation.

Leaders are asking harder operational questions.

Which systems can genuinely be trusted?

Which decisions should remain human?

What governance structures are required before autonomy scales further into the enterprise?

This is a good thing: it shows the industry (and the technology and its applications) is ready to move on from failed experiments and are not afraid to buck hype.

In fact, one Group CISO around the table described fielding more than 80 generative and agentic AI minimum viable products across the organisation. They decided to scale none of them. The point was not that the technology had failed. Rather, there is growing executive recognition that experimentation is easy whereas operational trust is difficult.

Another part of the debate focused on build versus buy (in terms of AI systems) and leaders seem genuinely stuck. Many organisations can produce impressive pilots; far fewer are confident enough in the underlying governance, data quality, accountability structures or commercial logic to deploy these systems at enterprise scale.

Why enterprise AI pilots fail to scale

This is partly because AI’s impact continues to be jagged.

The reality differs significantly by sector, geography, regulatory environment and organisational maturity. In some companies, as we heard from Tredence, intelligent agents are already automating routine financial workflows, orchestrating customer engagement or helping software teams remediate issues autonomously. Elsewhere, however, organisations remain stuck trying to reconcile fragmented data estates and inconsistent governance policies.

This goes some way to explain how the same technology can create something as genuinely transformative as Alphafold, winning its creators Nobel Prizes, whereas in other sectors, dozens of pilots still get rejected.

As a result, the enterprise AI story of 2026 is less about technological possibility and more about institutional and cultural readiness.

That distinction matters because the latest wave of AI represents a deeper operational shift than many previous enterprise software cycles. Most earlier forms of enterprise AI largely assisted humans: recommending products, forecasting demand, summarising documents or identifying anomalies.

Agentic systems move closer to execution itself.

Inside organisations, this increasingly means systems capable of triggering workflows, escalating decisions, coordinating actions across applications and adapting dynamically to changing conditions. A supply chain platform rerouting inventory automatically in response to disruptions. A finance function handling reconciliation and exception management with minimal intervention. Customer experience systems adjusting interactions in real time according to behaviour, sentiment or commercial value.

The important nuance is that this is not full autonomy in the science-fiction sense often portrayed in popular debate. What enterprises are actually building are layered systems of delegated execution: AI operating within defined guardrails, escalation pathways, and confidence thresholds.

Yet even this more constrained version of autonomy raises difficult questions about how organisations themselves are structured, we heard.

For years, digital transformation focused largely on efficiency and visibility. Agentic AI introduces something more consequential: the prospect of software participating directly in operational coordination. The shift is subtle but important. Digitising workflows was a priority for the last decade; now leaders are expected to embed a proxy for intelligence inside them.

Trust and governance in autonomous systems

This helps explain why so many AI initiatives struggle to progress beyond experimentation.

The obstacles are rarely confined to the models themselves. More often, intelligent agents expose weaknesses that already existed within organisations: fragmented systems, poor interoperability, inconsistent data lineage, unclear ownership of processes and governance frameworks that were designed for slower, more deterministic technologies. That last point is especially pertinent for any CFO reviewing return-on-investment metrics.

In many boardrooms, trust—within the technology, and between C-suite leaders—is becoming a defining intangible in enterprise AI adoption, therefore.

Technology teams may demonstrate increasingly capable systems, but boards and executive committees remain wary of operational fragility given many are held personally accountable in a fractious regulatory environment.

What’s more, the speed with which AI systems are being integrated into workflows sometimes runs ahead of the maturity of the surrounding organisation. Architecture, governance, and oversight structures evolve more slowly than deployment ambitions, risk leaders are at pains to remind their peers.

This tension is becoming increasingly visible.

Directors hear competitors describing accelerated AI deployments and fear strategic drift if their own organisations move too cautiously. Yet many CIOs, CISOs, and Chief Data Officers privately describe a widening gap between executive ambition and operational preparedness, as well as acknowledgement of the so-called prisoner’s dilemma: strategy as FOMO, or fear-of-missing-out.

The result is a new kind of executive pressure: not necessarily fear that AI will fail, but fear that organisations may scale systems they do not yet fully understand.

This is both a signal for leaders to check their visions, and an opportunity to focus on the actual competitive differentiators a company holds: data, and talent.

The role of data and talent in agentic AI

Agentic AI increasingly debates rarely question whether the technology works outside of hallucinations, but instead hones in on where human judgement should continue to reside.

Frontier AI organisations such as Anthropic segment the technology into two broad categories: narrow, and general. Both categories are scaled one to five in terms of maturity and veracity, five being the most mature. Mature but narrow case studies are performing best, we heard, when leaders pick a very specific, bounded problem and solve it well. The data is contained and the teams involved, secured, with good domain knowledge. The problem is also typically better defined, with a simpler categorisation of success or failure.

General AI, however, is still too broad in its application to be well understood, and immature to make an impact.

The future is unlikely to involve entirely autonomous enterprises soon, therefore. Far more plausible is the emergence of bounded autonomy: systems operating independently within carefully defined parameters, with humans retained for escalation, oversight and high-consequence decision-making.

In practice, this means governance becoming a design principle rather than a compliance afterthought. Organisations are increasingly focused on auditability, observability, reversibility of actions and clearer confidence thresholds for autonomous systems. Human-in-the-loop are evolving into something more nuanced: human-on-the-loop oversight, where intervention occurs selectively rather than continuously.

Underlying all of this is a renewed focus on data.

For companies pursuing agentic AI seriously, the conversation quickly returned to foundational questions around information architecture, interoperability and trust. Intelligent agents are only as effective as the quality, accessibility, and governance of the data environments in which they operate.

This is partly why firms such as Tredence and Databricks increasingly frame AI adoption less as a standalone tooling exercise and more as an enterprise data and operational challenge. The emphasis is shifting towards unified data environments, scalable governance, and production-grade AI systems capable of operating reliably across complex organisations.

But the data conversation itself is also evolving.

In some sectors, particularly media, publishing, and knowledge-intensive industries, leaders are beginning to rethink what constitutes strategic data in the first place. Customer data remains valuable, but many organisations are reconsidering the importance of proprietary content, institutional knowledge, and internal expertise as competitive AI assets.

In effect, some enterprises are discovering that the AI race is simultaneously becoming a race to organise, govern, and operationalise their own intellectual capital.

That may ultimately prove the more durable competitive advantage.

Closing thoughts

For decades, enterprise software largely focused on digitising processes. Agentic AI marks the beginning of something different: software proactively acting inside operational decision-making itself.

The implications are not arriving evenly. Adoption is varying by sector, regulation, organisational maturity, and cultural tolerance for risk. Many more pilots will fail. Some organisations will move too quickly; others may become paralysed by caution. The goldilocks zone of careful agility is one where CEOs are jockeying for position.

But the broader direction of travel is becoming clearer.

The defining enterprise question of the next few years may not be what AI systems are capable of doing per se, but how institutions reimagine what is possible with agents acting on their own behalf to get it done.

FAQs

1. Are enterprises already deploying agentic AI?

Yes, although adoption remains uneven. Some organisations are already using intelligent agents to automate financial workflows, optimise supply chains, orchestrate customer engagement, or support IT and engineering operations. However, many enterprises remain in pilot phases as they address governance, trust, and data challenges.

2. Why are so many AI pilots failing to scale?

Many organisations are discovering that the primary challenge is not the AI model itself, but institutional readiness. Common barriers include fragmented data estates, weak governance structures, poor interoperability between systems, unclear accountability, and limited organisational trust in autonomous processes.

3. Why is trust becoming such an important issue in enterprise AI?

Trust has emerged as a defining factor because organisations are embedding AI systems directly into operational workflows. Boards and executives increasingly want confidence not only in model performance, but also in the reliability of underlying data, governance processes, monitoring systems, and accountability structures.

4. What role does data play in successful agentic AI deployment?

Data quality, accessibility, governance, and interoperability are foundational. Intelligent agents are only as effective as the environments in which they operate. Many organisations are therefore investing in unified data architectures, scalable governance frameworks, and production-grade AI infrastructure before attempting to scale agentic systems.

5. What should executives focus on before scaling agentic AI?

Leaders should focus on governance, trusted data foundations, interoperability, observability, accountability, and clear operational boundaries for autonomous systems. Many successful deployments begin with narrow, well-defined use cases where outcomes can be measured and controlled before broader enterprise rollout.